移动机器人导航技术是智能机器人研究领域的一个重要方向，也是智能机器人的一项关键技术。近年来，随着图像处理技术及计算机处理能力的飞速发展，基于视觉信息的移动机器人导航是当前移动机器人研究的一个焦点。同人类一样，视觉也是机器人获取周围环境信息最重要的来源，因此，研究在复杂环境下的煤矿救援机器人的视觉系统具有极其重要的意义。 首先，在分析煤矿救援机器人应用环境的基础上，比较了机器人常用传感器的优缺点，确定了采用超声传感器、红外传感器和摄像机三种传感器构建煤矿救援机器人视觉系统的方案。利用红外传感器和超声传感器测距信息进行相互补充，能够有效地获取机器人附近障碍物的距离信息；利用摄像机能够获取障碍物的几何形状信息。 其次，以研祥嵌入式EC5-1717CLDNA计算机为上位机，以ATMEGA8单片机为下位机，构建了视觉系统数据采集与处理系统。利用单片机采集超声传感器、红外传感器测得的距离信息，并进行预处理，通过RS232串口将数据传送至上位机进行分析处理；基于CMOS图像传感器的工业摄像机捕捉到的数字图像信号通过USB2.0接口传输至上位机进行处理和识别。采用最小二乘法对超声传感器的测量值进行拟合，得到超声传感器较为精确的测距数学模型。采用均值法对红外传感器采集的数据进行滤波，提高了信噪比。以VisualC++6.0为软件开发平台，利用摄像机提供的接口函数实现了图像的实时采集。 最后，进行了实验研究。实验结果表明，本文提出的煤矿救援机器人方案可行，构建的检测系统结构紧凑，性能稳定，采集的距离信息、图像信息可靠，具有较高的精度，为机器人导航奠定了坚实的基础。

The navigation technology of mobile robot is an important research direction and a key technology of intelligent robot. In the last few years, along with rapid development of the image process technology and the computer operation ability, the vision-based navigation becomes one of the hot topics in robot field. Just like humans, vision is also the most important source for robot to know the world around. So，it is very important to research on the vision system of the coal mine rescue robot in a complex environment. Firstly, comparing the advantages and disadvantages of common sensor on the basis of the analysis of the environment of the robot, the project of the vision system by using ultrasonic ranging sensor, infrared sensors and camcorder is made sure. The distance information between the robot and obstructions is obtained by using ultrasonic and infrared sensors, and the obstruction’s geometry information is acquired by the camcorder. Secondly, a data collection and processing system of the visual system is built by YanXiang EC5-1717CLDNA Embedded-Computer as PC and ATMEGA8 SCM as the crew. In the vision system, the signal of ultrasonic ranging sensor and infrared sensor are get and pretreated by use of the single chip, and the data is transmitted by RS232 and is processed by PC. Images of obstructions are collected by industry camera based on CMOS sensor, and are transmitted to PC by USB2.0, then are identified by the computer. The values of the ultrasonic sensors are fit by the least-square method, and an accurate math model is gained. For measuring values of infrared sensors, an mean value method is used to filter, the Signal-to-Noise is improved. By means of the VisualC++6.0 software, the real-time image acquisition is realized by using the interface function of the camcorder. Finally, the experimental study is completed. The results show that the project on the vision system of the coal mine rescue robot is feasible. The structure of the detection system is compact, the performance is stable. The distance and image information are reliable and accurate. The research has established a good foundation for navigation of the robot.